NodeJS, ML, K8s and Unethical Face Recognition
summary
Prototype connects Twitter API (rate-limited at 900 calls per 15 minutes) to NodeJs services, uses face-api.js and TensorFlow.js with SSD MobileNet v1 and tiny face detector, and achieves per-image detection ~99% then ~83% in hard angles and 98% without glasses; one-shot recognition works from a single profile photo but produces false matches. Sentiment npm scores tweets between -5 and 5 for a neutrality-weighted metric. Architecture is microservices on OpenShift with RabbitMQ queues, MongoDB caching, autoscaling containers for slow face processing, and clear ethical failure modes.
AI-written summary · about
chapters (11)
- 0:00Introduction to the project
- 1:25Twitter API data collection
- 7:03Rating people with ML
- 10:14Face recognition with face-api.js
- 19:17Sentiment analysis setup
- 21:01Microservices architecture
- 22:29Containerization basics
- 27:47Messaging queues with RabbitMQ
- 31:18Kubernetes orchestration
- 35:14Live demo and results
- 36:38Unethical machine learning concerns
related talks
38mNodeJS, ML, K8s and Unethical Face Recognition
53mNodeJS, ML, K8s and Unethical Face Recognition
01h 05mIntro to machine learning in Node
26mSelf-Portraits in JavaScript
54mIntroduction to machine learning in Node.js
21mLeveraging Python for Real-Time Image processing
59mMachine Learning and Big Data Magic with Node.js and Google Cloud
31mFace Recognition